moonshotai

Kimi K3

kimi-k3

Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work

ReasoningTool CallingAttachmentsStructured OutputOpen Weights

Context Window

1.0M

Max Output

131K

Temperature

No

Open Weights

Yes

Knowledge Cutoff

N/A

Released

2026-07-16

Last Updated

2026-07-16

Modalities

Input:TextImageVideo
→
Output:Text

Benchmarks

BenchmarkScoreMetricHarness
DeepSWE67.5resolve rateKimi Code
Terminal-Bench88.3accuracyKimi Code
FrontierSWE81.2dominance scoreKimi Code
Program Bench77.8scoreKimi Code
SWE Marathon42resolve rateClaude Code
GDPval-AA1668Elo-
AA-Briefcase1548Elo-
AutomationBench30.8success rate-
JobBench52.9score-
SpreadsheetBench34.8scoreClaude Code
BrowseComp91.2accuracy-
CharXiv Reasoning91.3accuracy-
ZeroBench41pass@5-

Available Providers (14)

ProviderInput /1MOutput /1MCache Read /1MCache Write /1MReasoningStatus
TokenGoTokenGoRp 53.145Rp 265.721Rp 5.315—toggle, effort—
NvidiaNvidiaGratisGratis——toggle, effort—
NanoGPTNanoGPTRp 35.430Rp 177.148Rp 3.543—effort—
ImpossiblImpossiblRp 53.145Rp 265.721Rp 5.315—✓—
OpenRouterOpenRouterRp 53.145Rp 265.721Rp 5.315—toggle, effort—
Cloudflare AI GatewayCloudflare AI GatewayRp 53.145Rp 265.721Rp 5.315—toggle, effort—
ZenMuxZenMuxRp 53.145Rp 265.721Rp 5.315—toggle, effort—
ArceeArceeRp 53.145Rp 265.721Rp 5.315—effort—
OfoxOfoxRp 53.145Rp 265.721Rp 5.315—toggle—
Vercel AI GatewayVercel AI GatewayRp 53.145Rp 265.721Rp 5.315—✓—
NovitaAINovitaAIRp 53.145Rp 265.721Rp 5.315—toggle, effort—
ai&ai&Rp 53.145Rp 221.434Rp 8.858—effort—
TensorXTensorXRp 53.145Rp 265.721Rp 13.287—toggle—
Kilo GatewayKilo GatewayRp 53.145Rp 265.721Rp 5.315—effort—
← Kembali ke Katalog